Accelerating Diffusion Models for Generative AI Applications with Silicon Photonics

TL;DR

This paper introduces a silicon photonics-based accelerator that significantly improves the energy efficiency and throughput of diffusion models used in generative AI, addressing the high computational demands of these models.

Contribution

The paper presents a novel silicon photonics accelerator specifically designed for diffusion models, achieving substantial improvements over existing electronic platforms.

Findings

At least 3x better energy efficiency

5.5x throughput improvement

Effective acceleration of diffusion models

Abstract

Diffusion models have revolutionized generative AI, with their inherent capacity to generate highly realistic state-of-the-art synthetic data. However, these models employ an iterative denoising process over computationally intensive layers such as UNets and attention mechanisms. This results in high inference energy on conventional electronic platforms, and thus, there is an emerging need to accelerate these models in a sustainable manner. To address this challenge, we present a novel silicon photonics-based accelerator for diffusion models. Experimental evaluations demonstrate that our photonic accelerator achieves at least 3x better energy efficiency and 5.5x throughput improvement compared to state-of-the-art diffusion model accelerators.